ATH-AI创新事业部-Qoder Cloud Agent研发工程师-Agent Infra方向
R&D engineer on Alibaba's Qoder Cloud Agents team, building a serverless, fully-managed agent runtime (harness) exposed as API/SDK: agent-loop orchestration, session/memory/tool systems, sandbox pooling, and high-concurrency distributed infra. Core tech: Go/Java/Python, Kubernetes, containers/sandboxes, Langchain/Spring AI.
- Compensation
- Not disclosed
- Location
- Hangzhou
- Work setup
- Not specified
- Seniority
- Not specified
- Employment
- Not specified
- Experience
- 2+ years
- Category
- Not specified
- Posted
Skills
Must have
- Bachelor's or Master's degree, CS-related major preferred; 2+ years of distributed systems R&D experience (relaxable for exceptional candidates)
- Proficient in at least one of Go, Java, Python; strong OOP, concurrent and network programming; can independently architect and ship mid-scale services
- Strong distributed/high-concurrency skills: Kubernetes, containers, sandbox tech, distributed consistency, event-driven design, cache tiering, rate limiting/quota, capacity stress testing, large-scale
- Harness design/architecture capability (core): design or refactor a harness from scratch; agent loop, prompt engineering, context engineering, multi-agent collaboration
- Deep familiarity with mainstream open-source and commercial harnesses/Agent frameworks: e.g. Codex, OpenCode, Hermes Agent, Langchain, AgentScope, Spring AI, AWS AgentCore, Anthropic Managed Agents
- Excellent communication and collaboration, passion for continuous learning, ability to adapt quickly to rapid AI changes and drive product innovation
Nice to have
- Sandbox/Serverless runtime, large-scale storage & file mounting, model inference & cost optimization, agent evaluation & SWE-Bench, AI coding or dev-productivity tooling experience
- Active in tech communities; contributed to open source or built your own agent framework
Full description
Qoder Cloud Agents是阿里巴巴Qoder 推出的下一代云端全托管 Agent 平台,把 Agent 运行时(harness)以 API / SDK 的形式开放给企业——平台负责多轮推理编排、沙箱执行、会话状态、记忆与工具接入,为客户提供Serverless形态的Agent云服务。一方面对外提供企业级Agent Infra平台,一方面也承担Qoder 产品家族的统一云端Agent底座职责。
1、负责云端 Agent Harness 核心链路:Session / Turn 状态机、agent loop 编排、事件流处理、长程任务持久执行;
2、负责高并发分布式 Agent Infra 底盘:沙箱池化与预热、冷启动优化、跨资源与跨地域调度、配额限流与弹性伸缩、容量规划与常态化压测,支撑海量Agent调用的稳定与成本可控;
3、负责上下文、记忆与工具体系:Memory Store 与 Dreaming机制、Skills 与 MCP 工具集成、上下文工程优化,持续降低单位成功任务的耗时与 Token消耗;
4、负责被集成交付面:Agent模板体系、IM 渠道、生态Connector、Webhook 与事件订阅、定时与批量任务、多语言 SDK、多模态交互,为企业提供更多开箱即用的能力,打造B2B2C的Agent平台。
1、本科或硕士以上学历,计算机相关专业优先,2 年以上分布式系统研发经验,能力突出者可放宽;
2、精通 Go、Java、Python 中至少一种,熟练掌握 OOP 与并发编程、网络编程模型,能独立负责中等规模服务的架构与落地;
3、分布式与高并发硬功底:熟悉 Kubernetes / 容器 / 沙箱技术,掌握分布式一致性、事件驱动、缓存分层、限流配额与容量压测,有大规模在线服务稳定性建设经验;
4、具备 Harness 设计架构能力(核心要求):能从零设计或重构一套 harness,熟悉agent loop、提示词工程、上下文工程、多Agent协同等技术;
5、熟悉主流商业与开源 Harness 并深入研究过数家,熟悉其应用场景和技术原理:包括开源Harness,如Deepseek Harness、Codex、OpenCode、Openclaw、Hermes Agent;Agent开源框架,如Langchain、Agentscope、Spring ai等技术;商业Agent Infra,如AWS AgentCore harness、Anthropic Managed Agents;
6、优先条件:沙箱 / Serverless 运行时、大规模存储与文件挂载、模型推理与成本优化、Agent 评测与 SWE-Bench、AI Coding 或研发效能工具经验;热衷技术社区,参与过开源建设或自建 Agent 框架;
7、优秀的沟通与协作能力,持续学习的热情,能快速响应AI领域的快速变化,驱动产品技术创新。